Observability

Everything here is on by default.

Logs

One JSON object per line on stderr, with trace_id and span_id when a span is active:

{"time":"2026-10-28T14:59:47.115+00:00","level":"info","message":"POST /orders 201 3.1ms","service":"shop","trace_id":"08d2…","span_id":"250e…","context":{"method":"POST","route":"/orders","status":201,"duration_ms":3.1,"request_id":"01a1…","client_ip":"10.0.0.7"}}

Config: log.level (default info; warning in tests), log.stream (default php://stderr). Inject Psr\Log\LoggerInterface.

Traces

• A server span per request (GET /orders/{id}), continuing an incoming W3C traceparent.
• A client span per outbound HTTP call, with traceparent propagated.
• A span per SQL statement (db.query, with db.query.text).
• Your own: $tracer->trace('pricing.quote', fn (Span $span) => ..., ['tenant' => $id]).

Export with ->withOtlpTracing() in app.php and tracing.otlp_endpoint (default http://127.0.0.1:4318/v1/traces). Spans are exported after the response is sent. tracing.sample_ratio samples new traces; incoming sampling decisions are honoured.

Metrics

GET /metrics in Prometheus text format. Built in:

Metric	Labels
http_requests_total	method, route, status, worker
http_request_duration_seconds (histogram)	method, route, worker
http_client_requests_total	host, outcome (ok, 4xx, 5xx, error, circuit_open, bulkhead_full)
http_client_duration_seconds (histogram)	host

Add your own with Metrics::increment() and Metrics::observe(). Each worker reports its own series; sum across worker in queries. Protect the endpoint with metrics.token (required in production by the boot check) or disable it with metrics.enabled = false.

Health

• GET /health/live: the worker is up (uptime, requests handled).
• GET /health/ready: runs every service implementing HealthCheck (the database adds one automatically) and returns 503 while a check fails or the worker is draining after SIGTERM.